Continuous feed optimisation: testing titles and attributes the way you test ads

Module 4 taught you that an ad is never finished. Neither is the feed: every title is an "ad" with its own CTR and conversion rate, every attribute a piece that can be measured and improved. This lesson closes the module with the method for cultivating the feed — the same continuous optimisation cycle, applied to the catalogue.

Measure by attribute, not just by product

The Shopping report lets you segment by feed attributes: brand, category, product type, custom label, item ID. That answers:

Custom labels are also the testing mechanism: tagging the products in an experiment with a label turns any feed change into a measurable segment.

The title test on a pilot category

1. Pick a category with volume (hundreds of clicks a month).
2. Rewrite the titles by template (lesson 2) for only half the products (alternating, or by a label), via a supplemental feed. Tag both halves with custom_label_4 = "A" / "B".
3. Keep bids, budget and structure still.
4. 3-4 weeks. Compare by segment: impressions, CTR, conversions, value, ROAS.
5. If B wins: roll the template out to the whole category and then to the next ones. If not: review the template (is the order the one people actually use?) and repeat.

It is not a perfect test (the products are not identical), but with dozens of products on each side and weeks of data, the signal is clear. Log every test: category, template, dates, result.

Other feed tests worth running

• Main image: product alone vs product in context (the product alone usually wins, but not always in home decor).
• The first 70 characters: brand first vs type first in an ambiguous category.
• product_highlight and product_detail filled in vs empty, on the long tail.
• Price: not a feed test, a business test, but the feed measures it (elasticity per product).

The monthly feed routine

1. Merchant Diagnostics: disapprovals and warnings (module 5).
2. Shopping search terms for the month (module 5, lesson 3): vocabulary that converts and is not in the titles → rules or a supplemental feed.
3. Performance by attribute: categories/brands that spend without selling → exclude, lower the target or improve the feed?
4. Tests in progress: read the results, roll out the winners.
5. Margin labels: update them with price/cost changes.
6. New catalogue arrivals: do they come in with a templated title, GTIN, category, images? If not, the rule or the onboarding process is failing.

One hour a month for a mid-sized catalogue. Beyond that, automate.

The complete circuit

Feed (title, attributes)
  → Search terms (what Google understands)
    → Real sales per product (what works)
      → Feed correction (titles, labels, exclusions)
        → Feed…

Every turn of the circuit improves the vocabulary, clears the noise and concentrates spend on what is profitable. It is the compound interest of module 4, applied to a catalogue: the shop that cultivates its feed pays less for each sale every month, without touching bids.

💡 Ninja trick: the Shopping Ninja closes the circuit every night: it reads search terms and real sales per product, proposes titles with the vocabulary that converts, delivers them as a supplemental spreadsheet, logs what was applied and measures the product's before/after in the following weeks. You decide the template and approve with the checkbox; the machine measures, proposes and remembers.

What you should remember

• Measure by attribute (feed segments); custom labels are your testing tool.
• Title test on a pilot category: half A/B, 3-4 weeks, nothing else touched, logged.
• Monthly routine: diagnostics, terms → titles, attributes that spend, tests, margins, new arrivals.
• The feed → terms → sales → feed circuit is continuous optimisation applied to the catalogue.

End of Module 6. Module 7 changes network: Display and remarketing in depth — audiences, exclusions and where the money goes.